Add example about std.parallelism

[wilzbach]

[thewilsonator]

Might want to initialise it to iota. Also since the point is to parallelise it maybe add sone timing code to show that it runs faster?

[CyberShadow]

Eh, this is a bit boring. More interesting would be to calculate and print an actual result, such as I don't know, like sum of all prime numbers up to a point. Needs some ideas. taskPool.reduce and parallel sorting are also more interesting.

[quickfur]

Maybe a benchmark to showcase the difference between parallel and non-parallel foreach.

I'm not too concerned about the example being "interesting" in the sense of doing something complicated / clever; the point of a frontpage example is to show off the language, not the cleverness of the example writer. I think the key with std.parallelism is how easy it is to turn your single-threaded loop into a parallel loop (provided, of course, that loop iterations don't depend on each other).

[quickfur]

Here's a rough example that shows the difference:

void main()
{
    import std.datetime, std.math, std.parallelism, std.stdio;

    auto logs = new double[10_000_000];
    auto numIterations = 1;

    // Linear init
    auto bm1 = benchmark!({
        foreach (i, ref elem; logs)
        {
            elem = log(1.0 + i);
        }
    })(numIterations);
    writefln("Linear init: %s msecs", bm1[0].msecs);

    // Parallel init
    auto bm2 = benchmark!({
        foreach (i, ref elem; parallel(logs))
        {
            elem = log(1.0 + i);
        }
    })(numIterations);
    writefln("Parallel init: %s msecs", bm2[0].msecs);
}

[quickfur]

It can probably use some trimming down to size to fit on the front page :D

[CyberShadow]

I'm not too concerned about the example being "interesting" in the sense of doing something complicated / clever; the point of a frontpage example is to show off the language, not the cleverness of the example writer.

Sure, but calculating a result to memory then throwing it away is not great either.

[quickfur]

OK, maybe parallel reduce might be more interesting? Like summing an array of 1,000,000 doubles.

[dlang-bot]

Thanks for your pull request, @wilzbach!

Bugzilla references

Your PR doesn't reference any Bugzilla issue.

If your PR contains non-trivial changes, please reference a Bugzilla issue or create a manual changelog.

[wilzbach]

It can probably use some trimming down to size to fit on the front page :D

Tried to trim it down a bit. We might want to play with the array sizes a bit.
To give you a rough feeling:

500_000 x 100:
Linear init: 1639 msecs
Parallel init: 655 msecs

300_000 x 100:
Linear init: 988 msecs
Parallel init: 341 msecs

100_000 x 100:
Linear init: 313 msecs
Parallel init: 118 msecs

So the performance of the DLang-Tour backend, isn't that bad ;-)

Run it yourself.

Sure, but calculating a result to memory then throwing it away is not great either.

Hmm this (iteration of a list in parallel) is actually my most common use of std.parallelism.
The main point to show here is how amazingly easy it is to do parallelism in D.

[wilzbach]

So ... imho just showing off how easy & beautiful parallel algorithms can be is a major selling point (and we now easy benchmark to prove the point).

@CyberShadow what do you think?

[CyberShadow]

Alright.

Lines wrap at the default size:

[wilzbach]

Interesting - you seem to use a different font on your FF:

[CyberShadow]

Oh yeah, monospace can mean anything I guess. Okay.

[CyberShadow]

on your FF

Not browser specific BTW

[wilzbach]

monospace can mean anything I guess. Okay.

Btw:

font-family: Consolas, "Bitstream Vera Sans Mono", "Andale Mono", Monaco, "DejaVu Sans Mono", "Lucida Console", monospace;

I guess you don't have Consolas installed?

[wilzbach]

Interesting that the bot doesn't merge this ... will investigate.

[wilzbach]

I saw the following in the logs:

» 20 Jul 2017 02:50:40.247  163 <190>1 2017-07-20T00:50:39.742281+00:00 app web.1 - - [A3313FA1:00000000 dbg] Connection was closed by remote peer (fd 12).
» 20 Jul 2017 02:50:40.247  196 <190>1 2017-07-20T00:50:39.742322+00:00 app web.1 - - [A3313FA1:9C3A0DA1 dbg] Client didn't send the initial request in a timely manner. Closing connection.

...

 20 Jul 2017 02:52:42.330  137 <190>1 2017-07-20T00:52:41.815660+00:00 app web.1 - - [A3313FA1:9C3A0DA1 dbg] start task (fd 12).
» 20 Jul 2017 02:52:42.330  124 <190>1 2017-07-20T00:52:41.815686+00:00 app web.1 - - [A3313FA1:00000000 dbg] FD: -1
» 20 Jul 2017 02:52:42.330  164 <190>1 2017-07-20T00:52:41.815735+00:00 app web.1 - - [A3313FA1:00000000 dbg] Socket event on fd 12: 17 (33FCA88 vs 33FCA88)
» 20 Jul 2017 02:52:42.330  163 <190>1 2017-07-20T00:52:41.815770+00:00 app web.1 - - [A3313FA1:00000000 dbg] Connection was closed by remote peer (fd 12).
» 20 Jul 2017 02:52:42.330  196 <190>1 2017-07-20T00:52:41.815808+00:00 app web.1 - - [A3313FA1:9C3A0DA1 dbg] Client didn't send the initial request in a timely manner. Closing connection.
» 20 Jul 2017 02:52:42.330  135 <190>1 2017-07-20T00:52:41.815811+00:00 app web.1 - - [A3313FA1:9C3A0DA1 dbg] task out (fd 12).
» 20 Jul 2017 02:52:42.330  132 <190>1 2017-07-20T00:52:41.815817+00:00 app web.1 - - [A3313FA1:9C3A0DA1 dbg] task finished.

Will upgrade the bot to Vibe 0.8.0 later -maybe that helps.

[CyberShadow]

I guess you don't have Consolas installed?

That's a Microsoft font and IIRC not a core web font so that's probably it.

[wilzbach]

Will upgrade the bot to Vibe 0.8.0 later -maybe that helps.

FYI: dlang/dlang-bot#137 (deployed the upgrade of the bot to Vibe 0.8.0)